Mammography, a low-dose x ray technology for breast examination, is the primary screening method for early detection of breast cancer, significantly improving treatment success rates. Segmenting key structures in mammography images can enhance medical assessment by evaluating cancer risk and the quality of image acquisition. We introduce a series of data-centric strategies to enrich the training data for deep learning-based segmentation of landmark structures, such as the nipple, pectoral muscle, fibroglandular tissue, and fatty tissue. Our approach involves augmenting training samples through annotation-guided image intensity manipulation and style transfer, aiming for better generalization than standard training methods. These augmentations are applied in a balanced manner to ensure the model processes a diverse range of images from dierent vendor equipment while maintaining ecacy on the original data. We present extensive numerical and visual results demonstrating the superior generalization capabilities of our methods compared to standard training. This evaluation uses a large dataset of mammography images from various vendors. Additionally, we present complementary results showing both the strengths and limitations of our methods in dierent scenarios. The accuracy and robustness demonstrated in the experiments suggest that our method is well-suited for integration into clinical practice.